Grok4.20 officially launches four-tier inference mode, Elon Musk claims significant upgrade still pushing weekly

According to 1M AI News monitoring, xAI's Grok 4.20 officially concludes Beta testing today. Grok 4.20 introduces four inference modes: Auto, Fast, Expert, and Heavy. Auto automatically switches between Fast and Expert; Fast focuses on quick response; Expert engages in deep thought. xAI founder Elon Musk posted on X to collect user's first impressions of Grok 4.20, stating that "major upgrades are still being pushed out weekly."

In Artificial Analysis's Non-Delusion Rate test, Grok 4.20 Beta achieved a record-breaking 78% but received an overall intelligence score of 48, lagging behind Gemini 3.1 Pro and GPT-5.4 at 57. The API pricing is $2 for input, $6 per million tokens for output, supporting 2 million token super-long contexts.